A combined clinical and specific genes’ model to predict live birth for in vitro fertilization and embryo transfer patients

Abstract

Abstract Background: We aimed to develop an accurate model to predict live birth for patients receiving in vitro fertilization and embryo transfer (IVF-ET) treatment. Methods: This is a prospective nested case-control study. Women aged between 18 and 38 years, whose body mass index (BMI) were between the range of 18.5–24 kg/m2, who had an endometrium of ≥8 mm at the thickest were enrolled. All patients received IVF-ET treatment and were followed up until they had reproductive outcomes. Endometrial samples during the window of implantation (LH+ 6 to 9 days) were subjected to analyze specific endometrial receptivity genes’ expression using real-time PCR (RT-PCR). Patients were divided into live birth group and non-live birth group based on IVF-ET outcomes. Clinical signatures relevant to live birth were collected, analyzed, and used to establish a predictive model for live birth by univariate analysis (clinical model). Specific endometrial receptivity genes’ expression was analyzed, selected, and used to construct a predictive model for live birth by The Least Absolute Shrinkage and Selection Operator (LASSO) analysis (gene model). Finally, significant clinical factors and genes were used to construct a combined model for predicting live birth using multivariate logistical regression (combined model). Different models’ Area Under Curve (AUC) were compared to identify the most predictive model. Results: Thirty-nine patients were enrolled in the study, twenty-four patients had live births, fifteen did not. In univariate analysis, the odds of live birth for women with ovulation dysfunction was 4 times higher than that for women with other IVF-ET indications (OR=4.0, 95% CI: 1.125−8.910, P=0.018). Age, body mass index, duration of infertility, primary infertility, repeated implantation failure, antral follicle counting, ovarian sensitivity index, anti-Mullerian hormone, controlled ovarian hyperstimulation protocol and duration, total dose of FSH/hMG, number of oocytes retrieved, regiment of endometrial preparation, endometrium thickness before embryo transfer, type of embryo transferred were not associated with live birth (P>0.05). Only ovulation dysfunction was used to construct the clinical model and its AUC was 0.688. In lasso analysis, GAST, GPX3, THBS2 were found to promote the risk of live birth. AUCs for GAST, GPX3, THBS2 reached to 0.736, 0.672, and 0.678, respectively. The gene model was established based on these three genes and its AUC was 0.772. Ovulation dysfunction, GAST, GPX3, and THBS2 were finally used to construct the combined model, reaching the highest AUC (AUC=0.842). Conclusions: Compared to the single model, the combined clinical (Ovulation dysfunction) and specific genes’(GAST, GPX3, THBS2) model was more accurate to predict live birth for IVF-ET patients.